/*

* Copyright(c) 2019 Intel Corporation

* Copyright (c) 2016, Alliance for Open Media. All rights reserved

*

* This source code is subject to the terms of the BSD 2 Clause License and

* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License

* was not distributed with this source code in the LICENSE file, you can

* obtain it at https://www.aomedia.org/license/software-license. If the Alliance for Open

* Media Patent License 1.0 was not distributed with this source code in the

* PATENTS file, you can obtain it at https://www.aomedia.org/license/patent-license.

*/

#include <stdlib.h>

#include "aom_dsp_rtcd.h"

#include "definitions.h"

#include "enc_handle.h"

#include "cdef_process.h"

#include "cdef_copy.h"

#include "enc_dec_results.h"

#include "svt_threads.h"

#include "reference_object.h"

#include "enc_cdef.h"

#include "enc_dec_process.h"

#include "pic_buffer_desc.h"

#include "sequence_control_set.h"

#include "utility.h"

#include "pcs.h"

#include "resize.h"

#include "super_res.h"

static void set_unscaled_input_16bit(PictureControlSet* pcs) {

    EbPictureBufferDesc* input_pic  = pcs->ppcs->enhanced_unscaled_pic;

    EbPictureBufferDesc* output_pic = pcs->input_frame16bit;

    uint16_t             ss_x       = pcs->ppcs->scs->subsampling_x;

    uint16_t             ss_y       = pcs->ppcs->scs->subsampling_y;

    svt_aom_copy_buffer_info(input_pic, pcs->input_frame16bit);

    if (input_pic->bit_depth == EB_EIGHT_BIT) {

        svt_aom_convert_pic_8bit_to_16bit(input_pic, output_pic, ss_x, ss_y);

    } else {

        uint16_t* planes[3] = {

            (uint16_t*)output_pic->y_buffer, (uint16_t*)output_pic->u_buffer, (uint16_t*)output_pic->v_buffer};

        svt_aom_pack_2d_pic(input_pic, planes);

    }

}

static EbErrorType copy_recon_enc(SequenceControlSet* scs, EbPictureBufferDesc* recon_picture_src,

                                  EbPictureBufferDesc* recon_picture_dst, int num_planes, int skip_copy) {

    recon_picture_dst->border       = recon_picture_src->border;

    recon_picture_dst->width        = recon_picture_src->width;

    recon_picture_dst->height       = recon_picture_src->height;

    recon_picture_dst->max_width    = recon_picture_src->max_width;

    recon_picture_dst->max_height   = recon_picture_src->max_height;

    recon_picture_dst->bit_depth    = recon_picture_src->bit_depth;

    recon_picture_dst->color_format = recon_picture_src->color_format;

    recon_picture_dst->y_stride = recon_picture_src->y_stride;

    recon_picture_dst->u_stride = recon_picture_src->u_stride;

    recon_picture_dst->v_stride = recon_picture_src->v_stride;

    recon_picture_dst->luma_size   = recon_picture_src->luma_size;

    recon_picture_dst->chroma_size = recon_picture_src->chroma_size;

    recon_picture_dst->packed_flag = recon_picture_src->packed_flag;

    recon_picture_dst->y_stride_bit_inc = recon_picture_src->y_stride_bit_inc;

    recon_picture_dst->u_stride_bit_inc = recon_picture_src->u_stride_bit_inc;

    recon_picture_dst->v_stride_bit_inc = recon_picture_src->v_stride_bit_inc;

    recon_picture_dst->buffer_enable_mask = scs->seq_header.color_config.mono_chrome ? PICTURE_BUFFER_DESC_LUMA_MASK

                                                                                     : PICTURE_BUFFER_DESC_FULL_MASK;

    int      ss_x            = scs->subsampling_x;

    int      ss_y            = scs->subsampling_y;

    uint32_t bytes_per_pixel = scs->is_16bit_pipeline ? 2 : 1;

    // Get frame size to alloc

    uint32_t alloc_sz       = 0;

    uint32_t buffer_size[3] = {0};

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Y_FLAG) {

        alloc_sz += buffer_size[0] = recon_picture_dst->luma_size * bytes_per_pixel;

    }

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Cb_FLAG) {

        alloc_sz += buffer_size[1] = recon_picture_dst->chroma_size * bytes_per_pixel;

    }

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Cr_FLAG) {

        alloc_sz += buffer_size[2] = recon_picture_dst->chroma_size * bytes_per_pixel;

    }

    // Allocate the Picture Buffers (luma & chroma)

    EB_CALLOC_ALIGNED_ARRAY(recon_picture_dst->buffer_alloc, alloc_sz);

    recon_picture_dst->buffer_alloc_sz = alloc_sz;

    uint32_t assigned_space            = 0;

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Y_FLAG) {

        recon_picture_dst->y_buffer = recon_picture_dst->buffer_alloc +

            (recon_picture_dst->border + (recon_picture_dst->y_stride * recon_picture_dst->border)) * bytes_per_pixel;

        assigned_space += buffer_size[0];

    } else {

        recon_picture_dst->y_buffer = NULL;

    }

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Cb_FLAG) {

        recon_picture_dst->u_buffer = recon_picture_dst->buffer_alloc + assigned_space +

            ((recon_picture_dst->border >> ss_x) +

             (recon_picture_dst->u_stride * (recon_picture_dst->border >> ss_y))) *

                bytes_per_pixel;

        assigned_space += buffer_size[1];

    } else {

        recon_picture_dst->u_buffer = NULL;

    }

    if (recon_picture_dst->buffer_enable_mask & PICTURE_BUFFER_DESC_Cr_FLAG) {

        recon_picture_dst->v_buffer = recon_picture_dst->buffer_alloc + assigned_space +

            ((recon_picture_dst->border >> ss_x) +

             (recon_picture_dst->v_stride * (recon_picture_dst->border >> ss_y))) *

                bytes_per_pixel;

        assigned_space += buffer_size[2];

    } else {

        recon_picture_dst->v_buffer = NULL;

    }

    assert(assigned_space == alloc_sz);

    int use_highbd = scs->is_16bit_pipeline;

    if (!skip_copy) {

        assert(num_planes <= MAX_PLANES);

        for (int plane = 0; plane < num_planes; ++plane) {

            uint8_t *src_buf, *dst_buf;

            int32_t  src_stride, dst_stride;

            int sub_x = plane ? scs->subsampling_x : 0;

            int sub_y = plane ? scs->subsampling_y : 0;

            src_buf    = recon_picture_src->buffer[plane];

            src_stride = recon_picture_src->stride[plane];

            dst_buf    = recon_picture_dst->buffer[plane];

            dst_stride = recon_picture_dst->stride[plane];

            int height = ((recon_picture_src->height + sub_y) >> sub_y);

            for (int row = 0; row < height; ++row) {

                svt_memcpy(

                    dst_buf, src_buf, ((recon_picture_src->width + sub_x) >> sub_x) * sizeof(*src_buf) << use_highbd);

                src_buf += src_stride << use_highbd;

                dst_buf += dst_stride << use_highbd;

            }

        }

    }

    return EB_ErrorNone;

}

static void svt_av1_superres_upscale_frame(struct Av1Common* cm, PictureControlSet* pcs, SequenceControlSet* scs) {

    // Set these parameters for testing since they are not correctly populated yet

    EbPictureBufferDesc* recon_ptr;

    bool is_16bit = scs->is_16bit_pipeline;

    svt_aom_get_recon_pic(pcs, &recon_ptr, is_16bit);

    uint16_t  ss_x       = scs->subsampling_x;

    uint16_t  ss_y       = scs->subsampling_y;

    const int num_planes = scs->seq_header.color_config.mono_chrome ? 1 : MAX_PLANES;

    EbPictureBufferDesc  recon_pic_temp;

    EbPictureBufferDesc* ps_recon_pic_temp;

    ps_recon_pic_temp = &recon_pic_temp;

    EbErrorType return_error = copy_recon_enc(scs, recon_ptr, ps_recon_pic_temp, num_planes, 0);

    if (return_error != EB_ErrorNone) {

        ps_recon_pic_temp = NULL;

        assert(0);

    }

    EbPictureBufferDesc* src = ps_recon_pic_temp;

    EbPictureBufferDesc* dst = recon_ptr;

    // get the bit-depth from the encoder config instead of from the recon ptr

    int bit_depth = scs->static_config.encoder_bit_depth;

    assert(num_planes <= MAX_PLANES);

    for (int plane = 0; plane < num_planes; ++plane) {

        uint8_t *src_buf, *dst_buf;

        int32_t  src_stride, dst_stride;

        int sub_x  = plane ? ss_x : 0;

        int sub_y  = plane ? ss_y : 0;

        src_buf    = src->buffer[plane];

        src_stride = src->stride[plane];

        dst_buf    = dst->buffer[plane];

        dst_stride = dst->stride[plane];

        svt_av1_upscale_normative_rows(cm,

                                       (const uint8_t*)src_buf,

                                       src_stride,

                                       dst_buf,

                                       dst_stride,

                                       (src->height + sub_y) >> sub_y,

                                       sub_x,

                                       bit_depth,

                                       is_16bit);

    }

    // free the memory

    EB_FREE_ALIGNED_ARRAY(ps_recon_pic_temp->buffer_alloc);

}

/**************************************

 * Cdef Context

 **************************************/

typedef struct CdefContext {

    EbFifo* cdef_input_fifo_ptr;

    EbFifo* cdef_output_fifo_ptr;

    // Hoisted per-thread CDEF search scratch (were large on-stack arrays).

    uint16_t* tmp_dst; // 1 << (MAX_SB_SIZE_LOG2 * 2)

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

    uint16_t* inbuf; // CDEF_INBUF_SIZE (16-bit sentinel buffer)

#endif

#if CDEF_8BITS_PATH

    uint8_t* inbuf8; // CDEF_INBUF_SIZE (8-bit native interior path)

#endif

} CdefContext;

static void cdef_context_dctor(EbPtr p) {

    EbThreadContext* thread_ctx = (EbThreadContext*)p;

    CdefContext*     obj        = (CdefContext*)thread_ctx->priv;

    EB_FREE_ALIGNED_ARRAY(obj->tmp_dst);

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

    EB_FREE_ALIGNED_ARRAY(obj->inbuf);

#endif

#if CDEF_8BITS_PATH

    EB_FREE_ALIGNED_ARRAY(obj->inbuf8);

#endif

    EB_FREE_ARRAY(obj);

}

/******************************************************

 * Cdef Context Constructor

 ******************************************************/

EbErrorType svt_aom_cdef_context_ctor(EbThreadContext* thread_ctx, const EbEncHandle* enc_handle_ptr, int index) {

    CdefContext* cdef_ctx;

    EB_CALLOC_ARRAY(cdef_ctx, 1);

    thread_ctx->priv  = cdef_ctx;

    thread_ctx->dctor = cdef_context_dctor;

    // Input/Output System Resource Manager FIFOs

    cdef_ctx->cdef_input_fifo_ptr  = svt_system_resource_get_consumer_fifo(enc_handle_ptr->dlf_results_resource_ptr,

                                                                          index);

    cdef_ctx->cdef_output_fifo_ptr = svt_system_resource_get_producer_fifo(enc_handle_ptr->cdef_results_resource_ptr,

                                                                           index);

    // Hoisted per-thread CDEF search scratch (previously large on-stack arrays).

    EB_MALLOC_ALIGNED_ARRAY(cdef_ctx->tmp_dst, 1 << (MAX_SB_SIZE_LOG2 * 2));

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

    EB_MALLOC_ALIGNED_ARRAY(cdef_ctx->inbuf, CDEF_INBUF_SIZE);

#endif

#if CDEF_8BITS_PATH

    EB_MALLOC_ALIGNED_ARRAY(cdef_ctx->inbuf8, CDEF_INBUF_SIZE);

#endif

    return EB_ErrorNone;

}

#define default_mse_uv 1040400

static uint64_t compute_cdef_dist(const EbByte dst, int32_t doffset, int32_t dstride, const uint8_t* src,

                                  const CdefList* dlist, int32_t cdef_count, BlockSize bsize, int32_t coeff_shift,

                                  uint8_t subsampling_factor, bool is_16bit) {

    uint64_t curr_mse = 0;

    if (is_16bit) {

        curr_mse = svt_compute_cdef_dist_16bit(((uint16_t*)dst) + doffset,

                                               dstride,

                                               (uint16_t*)src,

                                               dlist,

                                               cdef_count,

                                               bsize,

                                               coeff_shift,

                                               subsampling_factor);

    } else {

        curr_mse = svt_compute_cdef_dist_8bit(

            dst + doffset, dstride, src, dlist, cdef_count, bsize, coeff_shift, subsampling_factor);

    }

    return curr_mse;

}

/* Search for the best filter strength pair for each 64x64 filter block.

 *

 * For each 64x64 filter block and each plane, search the allowable filter strength pairs.

 * Call cdef_filter_fb() to perform filtering, then compute the MSE for each pair.

*/

static void cdef_seg_search(CdefContext* ctx, PictureControlSet* pcs, SequenceControlSet* scs, uint32_t segment_index) {

    PictureParentControlSet* ppcs    = pcs->ppcs;

    FrameHeader*             frm_hdr = &ppcs->frm_hdr;

    Av1Common*               cm      = ppcs->av1_cm;

    uint32_t                 x_seg_idx;

    uint32_t                 y_seg_idx;

    const uint32_t           b64_pic_width  = (ppcs->aligned_width + 64 - 1) / 64;

    const uint32_t           b64_pic_height = (ppcs->aligned_height + 64 - 1) / 64;

    SEGMENT_CONVERT_IDX_TO_XY(segment_index, x_seg_idx, y_seg_idx, pcs->cdef_segments_column_count);

    const uint32_t x_b64_start_idx = SEGMENT_START_IDX(x_seg_idx, b64_pic_width, pcs->cdef_segments_column_count);

    const uint32_t x_b64_end_idx   = SEGMENT_END_IDX(x_seg_idx, b64_pic_width, pcs->cdef_segments_column_count);

    const uint32_t y_b64_start_idx = SEGMENT_START_IDX(y_seg_idx, b64_pic_height, pcs->cdef_segments_row_count);

    const uint32_t y_b64_end_idx   = SEGMENT_END_IDX(y_seg_idx, b64_pic_height, pcs->cdef_segments_row_count);

    const int32_t       mi_rows                    = cm->mi_rows;

    const int32_t       mi_cols                    = cm->mi_cols;

    CdefSearchControls* cdef_ctrls                 = &ppcs->cdef_search_ctrls;

    const int           first_pass_fs_num          = cdef_ctrls->first_pass_fs_num;

    const int           default_second_pass_fs_num = cdef_ctrls->default_second_pass_fs_num;

    EbByte              src[3];

    EbByte              ref[3];

    int32_t             stride_src[3];

    int32_t             stride_ref[3];

    int32_t             plane_bsize[3];

    int32_t             mi_wide_l2[3];

    int32_t             mi_high_l2[3];

    int32_t             xdec[3];

    int32_t             ydec[3];

    int32_t             cdef_count;

    const int32_t       coeff_shift = AOMMAX(scs->static_config.encoder_bit_depth - 8, 0);

    const int32_t       nvfb        = (mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;

    const int32_t       nhfb        = (mi_cols + MI_SIZE_64X64 - 1) / MI_SIZE_64X64;

    const int32_t       damping     = 3 + (frm_hdr->quantization_params.base_q_idx >> 6);

    const int32_t       num_planes  = 3;

    const bool          sb64        = scs->seq_header.sb_size == BLOCK_64X64;

    CdefList            dlist_local[MI_SIZE_128X128 * MI_SIZE_128X128];

    CdefList*           dlist;

#if CONFIG_ENABLE_HIGH_BIT_DEPTH

    const bool is_16bit = scs->is_16bit_pipeline;

#else

    const bool is_16bit = false;

#endif

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

    int32_t   toff_prev  = CDEF_VBORDER;

    int32_t   loff_prev  = CDEF_HBORDER;

    int32_t   ysize_prev = (1 << MAX_SB_SIZE_LOG2) + 2 * CDEF_VBORDER;

    int32_t   xsize_prev = (1 << MAX_SB_SIZE_LOG2) + 2 * CDEF_HBORDER;

    uint16_t* inbuf      = ctx->inbuf; // 16-bit sentinel buffer (HBD / non-8bit path only)

    uint16_t* in         = inbuf + CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER;

#endif

#if CDEF_8BITS_PATH

    // 8-bit padded buffer for the native interior path (replaces the 8->16 widen

    // for fully-interior 8-bit fbs). Same CDEF_BSTRIDE layout as `in`.

    uint8_t* inbuf8 = ctx->inbuf8;

    uint8_t* in8    = inbuf8 + CDEF_VBORDER * CDEF_BSTRIDE + CDEF_HBORDER;

#endif

    // tmp_dst is uint16_t to accommodate high bit depth content; 8bit will treat it as a uint8_t

    // buffer and will not use half of the buffer

    uint16_t* tmp_dst = ctx->tmp_dst;

    EbPictureBufferDesc* input_pic = is_16bit ? pcs->input_frame16bit : ppcs->enhanced_pic;

    EbPictureBufferDesc* recon_pic;

    svt_aom_get_recon_pic(pcs, &recon_pic, is_16bit);

    for (int pli = 0; pli < num_planes; pli++) {

        const int subsampling_x = (pli == 0) ? 0 : 1;

        const int subsampling_y = (pli == 0) ? 0 : 1;

        xdec[pli]               = subsampling_x;

        ydec[pli]               = subsampling_y;

        // The checks are stubs for 4:2:2 and 4:4:4 support

        // cppcheck-suppress knownConditionTrueFalse

        plane_bsize[pli] = subsampling_y ? (subsampling_x ? BLOCK_4X4 : BLOCK_8X4)

                                         : (subsampling_x ? BLOCK_4X8 : BLOCK_8X8);

        mi_wide_l2[pli]  = MI_SIZE_LOG2 - subsampling_x;

        mi_high_l2[pli]  = MI_SIZE_LOG2 - subsampling_y;

        src[pli]         = pcs->cdef_input_recon[pli];

        ref[pli]         = pcs->cdef_input_source[pli];

        stride_src[pli]  = pli == 0 ? recon_pic->y_stride : (pli == 1 ? recon_pic->u_stride : recon_pic->v_stride);

        stride_ref[pli]  = pli == 0 ? input_pic->y_stride : (pli == 1 ? input_pic->u_stride : input_pic->v_stride);

    }

    // Loop over all filter blocks (64x64)

    for (uint32_t fbr = y_b64_start_idx; fbr < y_b64_end_idx; ++fbr) {

        for (uint32_t fbc = x_b64_start_idx; fbc < x_b64_end_idx; ++fbc) {

            int32_t           dirinit = 0;

            const uint32_t    lc      = MI_SIZE_64X64 * fbc;

            const uint32_t    lr      = MI_SIZE_64X64 * fbr;

            int               nhb     = AOMMIN(MI_SIZE_64X64, mi_cols - lc);

            int               nvb     = AOMMIN(MI_SIZE_64X64, mi_rows - lr);

            int               hb_step = 1; //these should be all time with 64x64 SBs

            int               vb_step = 1;

            BlockSize         bs      = BLOCK_64X64;

            const MbModeInfo* mbmi    = pcs->mi_grid_base[lr * cm->mi_stride + lc];

            const BlockSize   bsize   = mbmi->bsize;

            if (((fbc & 1) && (bsize == BLOCK_128X128 || bsize == BLOCK_128X64)) ||

                ((fbr & 1) && (bsize == BLOCK_128X128 || bsize == BLOCK_64X128))) {

                continue;

            }

            if (bsize == BLOCK_128X128 || bsize == BLOCK_128X64 || bsize == BLOCK_64X128) {

                bs = bsize;

            }

            if (bs == BLOCK_128X128 || bs == BLOCK_128X64) {

                nhb     = AOMMIN(MI_SIZE_128X128, cm->mi_cols - lc);

                hb_step = 2;

            }

            if (bs == BLOCK_128X128 || bs == BLOCK_64X128) {

                nvb     = AOMMIN(MI_SIZE_128X128, cm->mi_rows - lr);

                vb_step = 2;

            }

            const uint32_t fb_idx = fbr * nhfb + fbc;

            // For SB=64, compute the dlist straight into the per-fb cache so the apply can reuse it

            // (skips a second svt_sb_compute_cdef_list scan); otherwise use the local scratch.

            dlist      = sb64 ? pcs->cdef_fb_list[fb_idx].dlist : dlist_local;

            cdef_count = svt_sb_compute_cdef_list(pcs, cm, lr, lc, dlist, bs);

            if (sb64) {

                pcs->cdef_fb_list[fb_idx].cdef_count = cdef_count;

            }

            if (cdef_count == 0) {

                pcs->skip_cdef_seg[fb_idx] = 1;

                continue;

            }

            pcs->skip_cdef_seg[fb_idx] = 0;

            int32_t toff = CDEF_VBORDER * (fbr != 0);

            int32_t loff = CDEF_HBORDER * (fbc != 0);

            int32_t boff = CDEF_VBORDER * ((int32_t)fbr + vb_step < nvfb);

            int32_t roff = CDEF_HBORDER * ((int32_t)fbc + hb_step < nhfb);

            uint8_t (*dir)[CDEF_NBLOCKS][CDEF_NBLOCKS] = &pcs->cdef_dir_data[fb_idx].dir;

            int32_t (*var)[CDEF_NBLOCKS][CDEF_NBLOCKS] = &pcs->cdef_dir_data[fb_idx].var;

            for (int pli = 0; pli < num_planes; pli++) {

                int32_t ysize = (nvb << mi_high_l2[pli]) + boff + toff;

                int32_t xsize = (nhb << mi_wide_l2[pli]) + roff + loff;

#if CDEF_8BITS_PATH

                // 8-bit content (RTC): filter via the boundary-aware hybrid. Build the 8-bit in8

                // buffer DIRECTLY from recon (interior AND frame-edge fbs); off-frame halo is left

                // as garbage and masked geometrically by the bounded kernel -- no 16-bit sentinel.

                const bool native_8bit = !is_16bit;

#else

                const bool native_8bit = false;

#endif

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

                if (!native_8bit) {

                    /* We avoid filtering the pixels for which some of the pixels to

                   average are outside the frame. We could change the filter instead,

                   but it would add special cases for any future vectorization.

                   Avoid memset'ting when dirty rect is inside the new one.

                   TODO: this could be further optimized - fill out only borders, separate buffers for Y & UV */

                    bool need_to_reset = toff_prev > toff || loff_prev > loff || ysize < ysize_prev ||

                        xsize < xsize_prev;

                    if (need_to_reset) {

                        uint16_t* p = &in[(-toff_prev * CDEF_BSTRIDE - loff_prev)];

                        for (int r = 0; r < ysize_prev; r++) {

                            svt_memset(p, (uint8_t)CDEF_VERY_LARGE, sizeof(p[0]) * xsize_prev);

                            p += CDEF_BSTRIDE;

                        }

                    }

                    toff_prev  = toff;

                    loff_prev  = loff;

                    ysize_prev = ysize;

                    xsize_prev = xsize;

                    svt_aom_copy_sb8_16(&in[(-toff * CDEF_BSTRIDE - loff)],

                                        CDEF_BSTRIDE,

                                        src[pli],

                                        (lr << mi_high_l2[pli]) - toff,

                                        (lc << mi_wide_l2[pli]) - loff,

                                        stride_src[pli],

                                        ysize,

                                        xsize,

                                        is_16bit);

                }

#endif

#if CDEF_8BITS_PATH

                if (native_8bit) {

                    // The bounded 8-bit kernel reads only +/-CDEF_HALO around the body, so copy just a

                    // HALO-wide halo (not the HBORDER/VBORDER SIMD padding). Off-frame sides have 0 halo;

                    // those taps are masked geometrically. svt_cdef_copy_rect8 dispatches on the (now

                    // apply-matching) width to a constant-size memcpy.

                    const int32_t toff8  = CDEF_HALO * (fbr != 0);

                    const int32_t loff8  = CDEF_HALO * (fbc != 0);

                    const int32_t boff8  = CDEF_HALO * ((int32_t)fbr + vb_step < nvfb);

                    const int32_t roff8  = CDEF_HALO * ((int32_t)fbc + hb_step < nhfb);

                    const int32_t ysize8 = (nvb << mi_high_l2[pli]) + boff8 + toff8;

                    const int32_t xsize8 = (nhb << mi_wide_l2[pli]) + roff8 + loff8;

                    svt_cdef_copy_rect8(&in8[(-toff8 * CDEF_BSTRIDE - loff8)],

                                        CDEF_BSTRIDE,

                                        src[pli],

                                        (lr << mi_high_l2[pli]) - toff8,

                                        (lc << mi_wide_l2[pli]) - loff8,

                                        stride_src[pli],

                                        ysize8,

                                        xsize8);

                }

#endif

                uint8_t subsampling_factor = cdef_ctrls->subsampling_factor;

                /*

                Cap the subsampling for certain block sizes.

                The intrinsics process several lines simultaneously, so blocks can only be subsampled

                a finite amount before there is no more speed gain.  If the space between processed lines

                is too large, the intrinsics will begin accessing memory outside the block.

                */

                switch (plane_bsize[pli]) {

                case BLOCK_8X8:

                    subsampling_factor = MIN(subsampling_factor, 4);

                    break;

                case BLOCK_8X4:

                case BLOCK_4X8:

                    subsampling_factor = MIN(subsampling_factor, 2);

                    break;

                case BLOCK_4X4:

                    subsampling_factor = MIN(subsampling_factor, 1);

                    break;

                }

                /* first cdef stage

                 * Perform the pri_filter strength search for the current sub_block

                 */

                for (int gi = 0; gi < first_pass_fs_num; gi++) {

                    // Check if chroma filter is set to be tested

                    if (pli && (cdef_ctrls->default_first_pass_fs_uv[gi] == -1)) {

                        pcs->mse_seg[1][fb_idx][gi] = default_mse_uv * 64;

                        continue;

                    }

#if CDEF_8BITS_PATH

                    if (native_8bit) {

                        svt_cdef_filter_fb_lbd((uint8_t*)tmp_dst,

                                               0,

                                               in8,

                                               toff == 0,

                                               loff == 0,

                                               boff == 0,

                                               roff == 0,

                                               nvb << mi_high_l2[pli],

                                               nhb << mi_wide_l2[pli],

                                               xdec[pli],

                                               ydec[pli],

                                               *dir,

                                               &dirinit,

                                               *var,

                                               pli,

                                               dlist,

                                               cdef_count,

                                               cdef_ctrls->default_first_pass_fs[gi],

                                               damping,

                                               coeff_shift,

                                               subsampling_factor);

                    } // native_8bit

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

                    else

#endif

#endif

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

                        svt_cdef_filter_fb(is_16bit ? NULL : (uint8_t*)tmp_dst,

                                           is_16bit ? tmp_dst : NULL,

                                           0,

                                           in,

                                           xdec[pli],

                                           ydec[pli],

                                           *dir,

                                           &dirinit,

                                           *var,

                                           pli,

                                           dlist,

                                           cdef_count,

                                           cdef_ctrls->default_first_pass_fs[gi],

                                           damping,

                                           coeff_shift,

                                           subsampling_factor);

#endif

                    uint64_t curr_mse = compute_cdef_dist(

                        ref[pli],

                        (lr << mi_high_l2[pli]) * stride_ref[pli] + (lc << mi_wide_l2[pli]),

                        stride_ref[pli],

                        (uint8_t*)tmp_dst,

                        dlist,

                        cdef_count,

                        (BlockSize)plane_bsize[pli],

                        coeff_shift,

                        subsampling_factor,

                        is_16bit);

                    if (pli < 2) {

                        pcs->mse_seg[pli][fb_idx][gi] = curr_mse * subsampling_factor;

                    } else {

                        pcs->mse_seg[1][fb_idx][gi] += (curr_mse * subsampling_factor);

                    }

                }

                /* second cdef stage

                 * Perform the sec_filter strength search for the current sub_block

                 */

                for (int gi = first_pass_fs_num; gi < first_pass_fs_num + default_second_pass_fs_num; gi++) {

                    // Check if chroma filter is set to be tested

                    if (pli && (cdef_ctrls->default_second_pass_fs_uv[gi - first_pass_fs_num] == -1)) {

                        pcs->mse_seg[1][fb_idx][gi] = default_mse_uv * 64;

                        continue;

                    }

#if CDEF_8BITS_PATH

                    if (native_8bit) {

                        svt_cdef_filter_fb_lbd((uint8_t*)tmp_dst,

                                               0,

                                               in8,

                                               toff == 0,

                                               loff == 0,

                                               boff == 0,

                                               roff == 0,

                                               nvb << mi_high_l2[pli],

                                               nhb << mi_wide_l2[pli],

                                               xdec[pli],

                                               ydec[pli],

                                               *dir,

                                               &dirinit,

                                               *var,

                                               pli,

                                               dlist,

                                               cdef_count,

                                               cdef_ctrls->default_second_pass_fs[gi - first_pass_fs_num],

                                               damping,

                                               coeff_shift,

                                               subsampling_factor);

                    } // native_8bit

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

                    else

#endif

#endif

#if !CDEF_8BITS_PATH || CONFIG_ENABLE_HIGH_BIT_DEPTH

                        svt_cdef_filter_fb(is_16bit ? NULL : (uint8_t*)tmp_dst,

                                           is_16bit ? tmp_dst : NULL,

                                           0,

                                           in,

                                           xdec[pli],

                                           ydec[pli],

                                           *dir,

                                           &dirinit,

                                           *var,

                                           pli,

                                           dlist,

                                           cdef_count,

                                           cdef_ctrls->default_second_pass_fs[gi - first_pass_fs_num],

                                           damping,

                                           coeff_shift,

                                           subsampling_factor);

#endif

                    uint64_t curr_mse = compute_cdef_dist(

                        ref[pli],

                        (lr << mi_high_l2[pli]) * stride_ref[pli] + (lc << mi_wide_l2[pli]),

                        stride_ref[pli],

                        (uint8_t*)tmp_dst,

                        dlist,

                        cdef_count,

                        (BlockSize)plane_bsize[pli],

                        coeff_shift,

                        subsampling_factor,

                        is_16bit);

                    if (pli < 2) {

                        pcs->mse_seg[pli][fb_idx][gi] = curr_mse * subsampling_factor;

                    } else {

                        pcs->mse_seg[1][fb_idx][gi] += (curr_mse * subsampling_factor);

                    }

                }

            }

        }

    }

}

/******************************************************

 * CDEF Kernel

 ******************************************************/

EbErrorType svt_aom_cdef_kernel_iter(void* context) {

    // Context & SCS & PCS

    CdefContext*        context_ptr = (CdefContext*)context;

    PictureControlSet*  pcs;

    SequenceControlSet* scs;

    //// Input

    EbObjectWrapper* dlf_results_wrapper;

    DlfResults*      dlf_results;

    //// Output

    EbObjectWrapper* cdef_results_wrapper;

    FrameHeader* frm_hdr;

    // Get DLF Results

    EB_GET_FULL_OBJECT(context_ptr->cdef_input_fifo_ptr, &dlf_results_wrapper);

    dlf_results                   = (DlfResults*)dlf_results_wrapper->object_ptr;

    pcs                           = (PictureControlSet*)dlf_results->pcs_wrapper->object_ptr;

    PictureParentControlSet* ppcs = pcs->ppcs;

    scs                           = pcs->scs;

    bool       is_16bit                   = scs->is_16bit_pipeline;

    Av1Common* cm                         = pcs->ppcs->av1_cm;

    frm_hdr                               = &pcs->ppcs->frm_hdr;

    CdefSearchControls* cdef_search_ctrls = &pcs->ppcs->cdef_search_ctrls;

    if (!cdef_search_ctrls->use_reference_cdef_fs && !cdef_search_ctrls->use_qp_strength) {

        if (scs->seq_header.cdef_level && pcs->ppcs->cdef_level) {

            cdef_seg_search(context_ptr, pcs, scs, dlf_results->segment_index);

        }

    }

    //all seg based search is done. update total processed segments. if all done, finish the search and perfrom application.

    svt_block_on_mutex(pcs->cdef_search_mutex);

    pcs->tot_seg_searched_cdef++;

    if (pcs->tot_seg_searched_cdef == pcs->cdef_segments_total_count) {

        pcs->cdef_dist_dev = -1;

        if (scs->seq_header.cdef_level && pcs->ppcs->cdef_level) {

            finish_cdef_search(pcs);

            if (ppcs->enable_restoration || pcs->ppcs->is_ref || scs->static_config.recon_enabled) {

                // Do application iff there are non-zero filters

                if (frm_hdr->cdef_params.cdef_y_strength[0] != 0 || frm_hdr->cdef_params.cdef_uv_strength[0] != 0 ||

                    pcs->ppcs->nb_cdef_strengths != 1) {

                    svt_av1_cdef_frame(scs, pcs);

                }

            }

        } else {

            frm_hdr->cdef_params.cdef_bits           = 0;

            frm_hdr->cdef_params.cdef_y_strength[0]  = 0;

            pcs->ppcs->nb_cdef_strengths             = 1;

            frm_hdr->cdef_params.cdef_uv_strength[0] = 0;

        }

        if (pcs->ppcs->nb_cdef_strengths == 1 && frm_hdr->cdef_params.cdef_y_strength[0] == 0 &&

            frm_hdr->cdef_params.cdef_uv_strength[0] == 0) {

            pcs->cdef_dist_dev = 0;

        }

        //restoration prep

        bool is_lr = ppcs->enable_restoration && frm_hdr->allow_intrabc == 0;

        if (is_lr) {

            svt_av1_loop_restoration_save_boundary_lines(cm->frame_to_show, cm, 1);

            if (is_16bit) {

                set_unscaled_input_16bit(pcs);

            }

        }

        // ------- start: Normative upscaling - super-resolution tool

        if (frm_hdr->allow_intrabc == 0 && pcs->ppcs->frame_superres_enabled) {

            svt_av1_superres_upscale_frame(cm, pcs, scs);

        }

        if (scs->static_config.resize_mode != RESIZE_NONE) {

            EbPictureBufferDesc* recon = NULL;

            svt_aom_get_recon_pic(pcs, &recon, is_16bit);

            recon->width  = pcs->ppcs->render_width;

            recon->height = pcs->ppcs->render_height;

            if (is_lr) {

                EbPictureBufferDesc* input_pic = is_16bit ? pcs->input_frame16bit : pcs->ppcs->enhanced_unscaled_pic;

                svt_aom_assert_err(pcs->scaled_input_pic == NULL, "pcs_ptr->scaled_input_pic is not desctoried!");

                EbPictureBufferDesc* scaled_input_pic = NULL;

                // downscale input picture if recon is resized

                bool is_resized = recon->width != input_pic->width || recon->height != input_pic->height;

                if (is_resized) {

                    superres_params_type spr_params = {recon->width, recon->height, 0};

                    svt_aom_downscaled_source_buffer_desc_ctor(&scaled_input_pic, input_pic, spr_params);

                    svt_aom_resize_frame(input_pic,

                                         scaled_input_pic,

                                         scs->static_config.encoder_bit_depth,

                                         av1_num_planes(&scs->seq_header.color_config),

                                         scs->subsampling_x,

                                         scs->subsampling_y,

                                         input_pic->packed_flag,

                                         PICTURE_BUFFER_DESC_FULL_MASK,

                                         0); // is_2bcompress

                    pcs->scaled_input_pic = scaled_input_pic;

                }

            }

        }

        // ------- end: Normative upscaling - super-resolution tool

        pcs->rest_segments_column_count = scs->rest_segment_column_count;

        pcs->rest_segments_row_count    = scs->rest_segment_row_count;

        pcs->rest_segments_total_count  = (uint16_t)(pcs->rest_segments_column_count * pcs->rest_segments_row_count);

        pcs->tot_seg_searched_rest      = 0;

        pcs->ppcs->av1_cm->use_boundaries_in_rest_search = scs->use_boundaries_in_rest_search;

        pcs->rest_extend_flag[0]                         = false;

        pcs->rest_extend_flag[1]                         = false;

        pcs->rest_extend_flag[2]                         = false;

        uint32_t segment_index;

        for (segment_index = 0; segment_index < pcs->rest_segments_total_count; ++segment_index) {

            // Get Empty Cdef Results to Rest

            svt_get_empty_object(context_ptr->cdef_output_fifo_ptr, &cdef_results_wrapper);

            CdefResults* cdef_results   = (struct CdefResults*)cdef_results_wrapper->object_ptr;

            cdef_results->pcs_wrapper   = dlf_results->pcs_wrapper;

            cdef_results->segment_index = segment_index;

            // Post Cdef Results

            svt_post_full_object(cdef_results_wrapper);

        }

    }

    svt_release_mutex(pcs->cdef_search_mutex);

    // Release Dlf Results

    svt_release_object(dlf_results_wrapper);

    return EB_ErrorNone;

}

void* svt_aom_cdef_kernel(void* input_ptr) {

    EbThreadContext* thread_ctx = (EbThreadContext*)input_ptr;

    for (;;) {

        EbErrorType err = svt_aom_cdef_kernel_iter(thread_ctx->priv);

        if (err == EB_NoErrorFifoShutdown) {

            return NULL;

        }

    }

    return NULL;

}